Session 6 · Inside one day

Soil II: interflow and dry days

Session 5 produced runoff \(R\) on wet days. This session finishes the soil step. On wet days, \(R\) is divided between two paths to the river. On dry days, which at the Leaf River are most days, the soil and the groundwater have to supply the evaporative demand that the canopy could not.

Time about 30 minutesNeeds Session 5
Watch this session first: a 6-minute narrated video with English subtitles. The page below goes further and lets you try each step yourself.

You will be able to

Where we are: the rest of the soil step sends runoff along two paths and, on dry days, draws evaporation from the soil and the aquifer.

Two paths for runoff

Runoff can reach a stream over the ground surface (overland flow) or by moving sideways through the upper, more permeable layer of soil (interflow). How much the soil can carry sideways in a day depends on how permeable it is and how wet it is. dCREST expresses this as a daily interflow capacity:

\[ \tau \;=\; \frac{\mathit{SM}_c + \mathit{SM}'}{2\,\mathit{wm}} \cdot \mathit{fc} \cdot 24\ \text{h} \]

\(\mathit{SM}_c\): soil water before the storm; \(\mathit{SM}'\): after infiltration; their mean divided by wm is the soil's average wetness over the day, between 0 and 1. fc: interflow rate in mm per hour.

Runoff up to the capacity goes to interflow; anything beyond it, plus the impervious runoff \(P_i\) from Session 5, becomes overland flow:

\[ Q_{\text{int}} = \min(R, \tau) + X, \qquad Q_{\text{ov}} = R - \min(R, \tau) + P_i \]

\(X = \max(\mathit{SM} - \mathit{wm}, 0)\) is any soil water above capacity; it joins the interflow. In practice the soil never exceeds its capacity in dCREST, so \(X\) is a safeguard and stays at zero.

What the two paths mean in dCREST

The names suggest that overland flow is fast and interflow slow. In dCREST the timing is set by what happens next (Session 8): overland flow passes through the fast reservoir, which releases a fraction qc of its water each day, while interflow goes straight to the routing step. Changing fc therefore changes how much runoff is delayed by the reservoir.

At the Leaf River

Calibration gave the Leaf River a very small interflow rate, fc = 0.0185 mm/h. Even with a fully wet soil, \(\tau\) is at most \(0.0185 \times 24 = 0.44\) mm/day. On the storm of 27 August 1992, \(R\) was 29.5 mm and \(\tau\) only 0.26 mm. So 0.26 mm went to interflow and 29.3 mm to overland flow. Over 1989–1999 interflow carries only 3% of the model's runoff, overland flow through the fast reservoir 72%, and groundwater the rest. For this river, calibration effectively closed the interflow path and routed storm runoff through the fast reservoir.

Try itOpen the interflow path

The whole model is rerun from 1986 with your fc; the chart shows the winter of 1990–1991 (simulated in blue, observed as a dark line). With a large fc most runoff skips the fast reservoir and the peaks become sharper (and very slightly earlier). Watch the test score: is the calibrated value the best?

Dry days: where evaporation comes from

On a dry day all the water reaching the ground evaporates at once, and the soil must supply the remaining demand. A wet soil gives up water easily; a dry soil holds on to it more tightly, and plants close their pores. dCREST captures this by letting the soil supply the demand in proportion to how full it is:

\[ E_s \;=\; \min\!\Big( (D - E_c - L)\cdot\frac{\mathit{SM}_c}{\mathit{wm}},\;\; \mathit{SM}_c \Big) \]

Whatever the soil could not supply, \(U = D - E_c - L - E_s\), is partly met by the aquifer, again in proportion to how full it is:

\[ E_g \;=\; \min\!\Big( U\cdot\frac{G}{\mathit{hmaxaq}},\;\; G \Big) \]

This represents water that deep roots or vegetation along streams can reach from groundwater, and moisture rising from the water table. The demand left after that goes unmet: the catchment simply evaporates less than the atmosphere could take.

A hot, dry August day

On 15 August 1993 there was no rain, the canopy was empty, and the demand was \(D = 0.955 \times 5.05 = 4.82\) mm.

StepCalculationmm
soil wetness\(\mathit{SM}_c / \mathit{wm} = 162.3 / 235.2\)0.690
soil evaporation \(E_s\)\(4.82 \times 0.690\)3.33
unmet demand \(U\)\(4.82 - 3.33\)1.49
aquifer fullness\(G / \mathit{hmaxaq} = 95.7 / 106.9\)0.895
aquifer evaporation \(E_g\)\(1.49 \times 0.895\)1.34
total evaporation\(E_c + L + E_s + E_g\)4.66

The numbers shown are rounded; each result is computed from the unrounded values, so redoing a row by hand can differ in the last digit.

The catchment evaporated 4.66 of the 4.82 mm demanded. That same day the model's river flow was 0.347 mm/day against 0.349 observed, all of it from groundwater (Session 7).

Try itA dry day, by hand

The bar is the day's demand left after the canopy (with no rain). As the soil and the aquifer dry out, less of the demand is met: the catchment evaporates less than the atmosphere could take.

A summer of drying

Put the dry days together and the catchment slowly dries through summer: the soil and the aquifer both decline, and with them the evaporation they can supply. The chart shows summer 1993 from the full model.

DataSummer 1993, from the model

Top: soil water SM and aquifer storage G (mm). Bottom: evaporation from the canopy, soil and aquifer (mm/day) and rain. After each rain the canopy takes over; between rains the soil and aquifer supply less and less.

In the code: src/dcrest/kernels/soil.py (shortened)
# wet branch: split the runoff
temX = (SM_c + Wo_w) / WM / 2.0 * (FC * dt_hours)                  # tau
inter_w = torch.minimum(R_w, temX) + interflowExcess_w
over_w = R_w - torch.minimum(R_w, temX) + precipImperv

# dry branch: soil evaporation, then aquifer-supported evaporation
ExcessET = (adjPET - precip) * wetness       # wetness = SM_cd / WM with the default options
ExcessET_act = torch.minimum(ExcessET, SM_cd)                      # E_s
Wo_d = SM_cd - ExcessET_act
unmet = torch.clamp(adjPET - precip - ExcessET_act, min=0.0)       # U
resET_d = torch.minimum(unmet * gw / HMAX, gw)                     # E_g

wet = precip > adjPET
over = torch.where(wet, over_w, torch.zeros_like(SM0))
inter = torch.where(wet, inter_w, interflowExcess_d)
gw = gw - torch.where(wet, torch.zeros_like(SM0), resET_d)

Both branches are computed for every catchment, and torch.where picks the one that applies. Here precip is the water reaching the soil (\(L\) in the equations) and adjPET the demand left after the canopy.

In Python

A dry spell, day by day: start from the model's soil and aquifer at the beginning of 15 August 1993 and give the catchment 40 days of August demand and no rain.

Check your understanding

With fc = 2 mm/h and an average wetness of 0.5, what is the interflow capacity? If the day's runoff is 30 mm, how is it split?

\(\tau = 0.5 \times 2 \times 24 = 24\) mm/day. Interflow \(\min(30, 24) = 24\) mm, overland flow 6 mm (plus any impervious runoff).

In dCREST, which runoff path reaches the river faster, and why?

Interflow: it goes straight to the routing step, while overland flow first passes the fast reservoir, which releases only a fraction qc of its water per day.

On a dry day the soil is 30% full and the demand is 5 mm. How much does the soil evaporate?

\(5 \times 0.3 = 1.5\) mm (less than the soil water, so the minimum does not apply). The unmet 3.5 mm is then partly supplied by the aquifer.

Why does total evaporation fall during a long dry spell even if the atmosphere's demand stays the same?

Soil and aquifer evaporation are proportional to how full they are. As they drain, they supply less each day, so a growing part of the demand goes unmet.

What to remember